MCP servers — give your bot more tools

If your business (or a service you use) already has an MCP server — the Model Context Protocol that AI assistants use to call tools — your chatbot can use its tools too. Add the server once and pick which tools customers may use. Available on Growth and higher.

1. Add the server

   On the chatbot's Actions tab, click Connect MCP server. Enter a short name (e.g. store) and the server's URL (https://…/mcp). If it needs an API key, add it as a header, e.g. Authorization: Bearer ….

2. Review the tools

   We connect and list the server's tools. Tools the server marks as read-only are enabled directly; the others start with confirmation required.

3. Choose what customers can use

   Turn off every tool customers should never reach — admin, delete or bulk-change tools. Keep confirmation on for anything that changes data.

4. Protect customer data

   Turn on Only for verified customers if any tool reads or changes a customer's data, and identify signed-in customers.

5. Try it

   Ask a question in the Sandbox tab that should use a tool, then check the Activity list.

Requirements and limits

• The server must be reachable over the internet with the Streamable HTTP transport. Local (stdio) servers can't be used.
• Authentication is by fixed headers (API key or bearer token). Servers that only allow a browser OAuth sign-in aren't supported yet.
• Each tool result is shortened to 8,000 characters before the bot sees it.
• If you change the server's tools, press Refresh tools. A failed connection shows its error on the server's page.

For a verified customer every request carries the x-chatbotgen-identity token, so your server can check which customer the bot is acting for. See the developer guide.

See also

• Custom actions
• Identify signed-in customers